Hypothesis-Driven Autonomous Materials Synthesis with Multimodal LLM Agents
Abstract
Self-driving laboratories can explore synthesis conditions autonomously, but their decision-making layer is typically a black-box optimizer, and the output is a set of optimized samples, with the measurements reduced to predefined scalar objectives and the reasons behind success left unarticulated. Here we present SynAgent, a framework in which large language model agents operate an automated experimental system and maintain an explicit, revisable understanding of the synthesis process as the campaign's primary output. Starting with no predefined analysis pipeline, SynAgent adaptively generates analysis skills for newly acquired data and evolves this understanding through multimodal reasoning over experimental data such as X-ray diffraction patterns and electron micrographs. The evolution is guided by a verify-falsify scheme, in which the agent deliberately challenges its own hypotheses by testing conditions predicted to fail as well as those predicted to succeed. In a single campaign of 18 autonomous experiments using LiCoO2 (001) thin-film deposition as a testbed, SynAgent synthesized highly crystalline films and evolved an understanding of how the substrate temperature governs crystallization, discovering an abrupt threshold and a narrow optimal growth window at 650-690 °C. These results extend autonomous experimentation beyond optimized samples to testable, human-readable understanding.
Related
- Physics-Informed and Knowledge-Driven Generative AI for Autonomous Discovery of Porous Oxide Energy Materials: Opportunities and Challenges
- Autonomous Analysis of Thermoelectric Materials Big Data by Multi-Agentic AI and Construction of a Knowledge Base and Specialized AI
- Divergent strategies and convergent outcomes in autonomous materials discovery
- From Automated Simulation to Autonomous Discovery: A Hierarchical Framework for Agentic Computational Materials Science
- Mechanoluminescence in crystalline inorganic materials: local disorder and the elastic distortion hypothesis
BibTeX
@misc{takahara2026hypothesisdriven,
title = {{Hypothesis-Driven Autonomous Materials Synthesis with Multimodal LLM Agents}},
author = {Izumi Takahara and Kazunori Nishio and Akira Aiba and Shigeru Kobayashi and Takao Nakajima and Taro Hitosugi and Teruyasu Mizoguchi},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.18598},
eprint = {2609.18598},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.18598},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Hypothesis-Driven Autonomous Materials Synthesis with Multimodal LLM Agents
AU - Izumi Takahara
AU - Kazunori Nishio
AU - Akira Aiba
AU - Shigeru Kobayashi
AU - Takao Nakajima
AU - Taro Hitosugi
AU - Teruyasu Mizoguchi
PY - 2026
DA - 2026/09/16
PB - arXiv
DO - 10.48550/arxiv.2609.18598
UR - https://arxiv.org/abs/2609.18598
AB - Self-driving laboratories can explore synthesis conditions autonomously, but their decision-making layer is typically a black-box optimizer, and the output is a set of optimized samples, with the measurements reduced to predefined scalar objectives and the reasons behind success left unarticulated. Here we present SynAgent, a framework in which large language model agents operate an automated experimental system and maintain an explicit, revisable understanding of the synthesis process as the campaign's primary output. Starting with no predefined analysis pipeline, SynAgent adaptively generates analysis skills for newly acquired data and evolves this understanding through multimodal reasoning over experimental data such as X-ray diffraction patterns and electron micrographs. The evolution is guided by a verify-falsify scheme, in which the agent deliberately challenges its own hypotheses by testing conditions predicted to fail as well as those predicted to succeed. In a single campaign of 18 autonomous experiments using LiCoO2 (001) thin-film deposition as a testbed, SynAgent synthesized highly crystalline films and evolved an understanding of how the substrate temperature governs crystallization, discovering an abrupt threshold and a narrow optimal growth window at 650-690 °C. These results extend autonomous experimentation beyond optimized samples to testable, human-readable understanding.
ER -